Dat Tran

Vulnerability of speaker verification to voice mimicking

We consider mimicry, a simple technology form of attack requiring a low level of expertise, to investigate whether a speaker recognition system is vulnerable to mimicry by an impostor without using the assistance of any other technologies. Experiments on 138 speakers in the YOHO database and two people who played a role as imitators have shown that an impostor can attack the system if that impostor knows a registered speaker in the database who has very similar voice to the impostors voice.

Fuzzy entropy clustering

The well-known generalisation of hard c-means (HCM) clustering is fuzzy c-means (FCM) clustering where a weight exponent on each fuzzy membership is introduced as the degree of fuzziness. An alternative generalisation of HCM clustering is proposed in this paper. This is called fuzzy entropy (FE) clustering where a weight factor of the fuzzy entropy function is introduced as the degree of fuzzy entropy. The weight factor is similar to the weight exponent and has a physical interpretation. The noise clustering approach, the fuzzy covariance matrix and the fuzzy mixture weight are also proposed. Moreover, we can show Gaussian mixture clustering is regarded as a special case of FE clustering. Some illustrative examples are performed on the Butterfly and Iris data.

Senior health monitoring using Kinect

This paper presents a new senior health monitoring system using the Kinect device to monitor elderly people and detect when they are likely to fall by measuring their gait, and analyzing change in posture when they change from sitting to standing or vice versa. Support vector machine is used to analyze the gait and posture data obtained from the Kinect device. Several experiments were performed to evaluate the proposed system and experimental results as well as research experience on using the Kinect device will be presented.

Fuzzy hidden Markov models for speech and speaker recognition

The paper proposes a fuzzy approach to the hidden Markov model (HMM) method called the fuzzy HMM for speech and speaker recognition. The fuzzy HMM algorithm is regarded as an application of the fuzzy expectation-maximisation (EM) algorithm to the Baum-Welch algorithm in the HMM. Speech and speaker recognition experiments using the Texas Instruments (TI46) speech data corpus show better results for fuzzy HMMs compared with conventional HMMs.

Password Entropy and Password Quality

Passwords are the first line of defense for many computerized systems. The quality of these passwords decides the security strength of these systems. Many studies advocate using password entropy as an indicator for password quality where lower entropy suggests a weaker or less secure password. However, a closer examination of this literature shows that password entropy is very loosely defined. In this paper, we first discuss the calculation of password entropy and explain why it is an inadequate indicator of password quality. We then establish a password quality assessment scheme: password quality indicator (PQI). The PQI of a password is a pair (D, L), where D is the Levenshteins editing distance of the password in relation to a dictionary of words and common mnemonics, and L is the effective password length. Finally, we propose to use PQI to prescribe the characteristics of good quality passwords.

Remote multimodal biometric authentication using bit priority-based fragile watermarking

We propose a new remote multimodal biometric authentication framework based on fragile watermarking for transferring multi-biometrics over networks to server for authentication. A facial image is used as a container to embed other numeric biometrics features. The proposed framework enhances security and reduces bandwidths. In order to reduce error rates from embedding numeric information, we also propose a new method to determine bit priority level in a bit sequence representing the numerical information to be embedded and combine with the current amplitude modulation watermarking method.

An optimal sphere and two large margins approach for novelty detection

We introduce a new model to deal with imbalanced data sets for novelty detection problems where the normal class of training data set can be majority or minority class. The key idea is to construct an optimal hypersphere such that the inside margin between the surface of this sphere and the normal data and the outside margin between that surface and the abnormal data are as large as possible. Depending on a specific real application of novelty detection, the two margins can be adjusted to achieve the best true positive and false positive rates. Experimental results on a number of data sets showed that the proposed model can provide better performance comparing with current models for novelty detection.

Emotion recognition using the emotiv EPOC device

Emotion plays an important role in the interaction between humans as emotion is fundamental to human experience, influencing cognition, perception, learning communication, and even rational decision-making. Therefore, studying emotion is indispensable. This paper aims at finding the relationships between EEG signals and human emotions based on emotion recognition experiments that are conducted using the commercial Emotiv EPOC headset to record EEG signals while participants are watching emotional movies. Alpha, beta, delta and theta bands filtered from the recorded EEG signals are used to train and evaluate classifiers with different learning techniques including Support Vector Machine, k-Nearest Neighbour, Naive Bayes and AdaBoost.M1. Our experimental results show that we can use the Emotiv headset for emotion recognition and that the AdaBoost.M1 technique and the theta band provide the highest recognition rates.

Testing voice mimicry with the YOHO speaker verification corpus

The aim of this paper is to determine how vulnerable a speaker verification system is to conscious effort by impostors to mimic a client of the system. The paper explores systematically how much closer an impostor can get to another speakers voice by repeated attempts. Experiments on 138 speakers in the YOHO database and six people who played a role as imitators showed a fact that professional linguists could successfully attack the system. Non-professional people could have a good chance if they know their closest speaker in the database.

A Security Architecture for e-Health Services

This paper presents an alternative way to secure communications in e-health. During the communication processes, users exchange different types of information with different levels of sensitivities. For example, communications between a doctor and a patient contain data of higher levels of sensitivities than communications between a social worker and a nurse. The different levels of the sensitivities of the information are secured by using different types of security processes. In this paper, these different communication types and different levels of data sensitivities in e-health are explained, the requirements for each type for communications are described and the use of the cryptography to secure the communication is discussed.